基于并行全连接卷积神经网络模型的室内可见光信道的最优灯源布局

Aiming at the problems of uneven distribution of indoor LED source, the problem of NLOS channel, noise interference in the environment, obstruction, interior border and irregular room layout failed to consider for the traditional Lambert model, an optimal light source layout scheme based on PFCNN (Parallel Fully Connected Convolutional Neural Network) model is proposed for indoor visible light positioning. The fingerprint database is constructed by collecting the light source information, such as the coordinate, power, orientation angle of the light source and the corresponding light intensity distribution of the receiving plane. The parameters for light intensity distribution flatness are measured using Monte Carlo method, and a fully connected neural network and parallel fully connected neural network are proposed to build an indoor visible light channel model, the light intensity flatness prediction model is constructed using the proposed PFCNN model, and optimal light source layout is realized by Mot-PSO-K-Means++ (k-means++ to optimize particle swarm optimization with momentum) algorithm. Simulation analysis shows that the accuracy of the parallel fully connected convolutional neural networks is improved by 84.69% compared with that of fully connected neural networks. In the 5m×5m×3m indoor space, the light intensity flatness reaches 92% for the 4-LED layout, and light intensity ranges from 340lx to 440lx. The light intensity flatness reaches 91% for the 12-LED layout, and light intensity ranges from 980lx to 1120lx. Therefore, the scheme has higher flatness and stronger applicability, can be applied to actual indoor visible light positioning, which provides a theoretical support for in-depth research of visible light positioning.